Distributed, Privacy Preserving, Payments Fraud Detection System

Abstract

Various aspects of the disclosure relate to fraud detection for real-time electronic transactions occurring in a plurality of payments systems. At onboarding of a regional or local payments computing system into a distributed privacy preservice payments fraud detection system, a client computing system trains a centralized model of a specific fraud detection architecture on its local dataset. Changes (e.g., gradient matrices) to the model are identified and communicated and communicates its gradients (matrices) to a centralized server. Upon receipt of gradients from one or more client computing systems, the centralized server aggregates the client gradients (periodically, continually, and the like) and sends the updated model to each client computing systems for use in the next federated round.

1 . A computing platform comprising:
 a processor; and   memory storing instructions that, when executed by the processor, cause the computing platform to:
 train an artificial intelligence/machine learning (AI/ML) model on a training data set; 
 communicate, via a network and to a plurality of payments computing systems, the trained model; 
 receive, via the network, a gradient update to a local model on a first payments computing system of the plurality of payments computing systems; 
 re-train the model based on the gradient update; and 
 communicate, via the network and to the plurality of payments computing systems, the re-trained model to each of the plurality of payments computing systems. 
   
     
     
         2 . The computing platform of  claim 1 , wherein the instructions cause the computing platform to aggregate gradient updates received from at least two payments computing systems of the payments computing systems. 
     
     
         3 . The computing platform of  claim 1 , wherein the instructions cause the computing platform to retrain the model based on aggregated gradient updates received from at least two payments computing systems of the payments computing systems. 
     
     
         4 . The computing platform of  claim 1 , wherein at least one payments system of the plurality of payments computing system communicates via an enterprise computing network local to the computing platform. 
     
     
         5 . The computing platform of  claim 1 , wherein at least one payments system of the plurality of payments computing system communicates via an external computing network communicatively coupled to the network and wherein the network is local to the computing platform. 
     
     
         6 . The computing platform of  claim 1 , wherein the first payments computing system operates using a first messaging protocol and a second computing payments system operates using a second messaging protocol. 
     
     
         7 . The computing platform of  claim 1 , wherein the first payments computing system operates using a standard messaging protocol and a second payments computing system operates using a proprietary messaging protocol. 
     
     
         8 . Non-transitory computer readable media storing instructions that, when executed by a processor, cause a fraud detection system to:
 train an artificial intelligence/machine learning (AI/ML) model on a training data set;   communicate, via a network and to a plurality of payments computing systems, the trained model;   receive, via the network, a gradient update to a local model on a first payments computing system of the plurality of payments computing systems;   re-train the model based on the gradient update; and   communicate, via the network and to the plurality of payments computing systems, the re-trained model to each of the plurality of payments computing systems.   
     
     
         9 . The non-transitory computer readable media of  claim 8 , wherein the instructions cause the fraud detection system to aggregate gradient updates received from at least two payments computing systems of the payments computing systems. 
     
     
         10 . The non-transitory computer readable media of  claim 8 , wherein the instructions cause the fraud detection system to retrain the model based on aggregated gradient updates received from at least two payments computing systems of the payments computing systems. 
     
     
         11 . The non-transitory computer readable media of  claim 8 , wherein at least one payments system of the plurality of payments computing system communicates via an enterprise computing network local to the fraud detection system. 
     
     
         12 . The non-transitory computer readable media of  claim 8 , wherein at least one payments system of the plurality of payments computing system communicates via an external computing network communicatively coupled to the network and wherein the network is local to the fraud detection system. 
     
     
         13 . The non-transitory computer readable media of  claim 8 , wherein the first payments computing system operates using a first messaging protocol and a second payments computing system operates using a second messaging protocol. 
     
     
         14 . The non-transitory computer readable media of  claim 8 , wherein the first payments computing system operates using a standard messaging protocol and a second payments computing system operates using a proprietary messaging protocol. 
     
     
         15 . A method comprising:
 training, by a computing platform, an artificial intelligence/machine learning (AI/ML) model on a training data set;   communicating, via a network and to a plurality of payments computing systems, the trained model;   receiving, via the network, a gradient update to a local model on a first payments computing system of the plurality of payments computing systems;   re-training the model based on the gradient update; and   communicating, via the network and to the plurality of payments computing systems, the re-trained model to each of the plurality of payments computing systems.   
     
     
         16 . The method of  claim 15 , further comprising aggregating gradient updates received from at least two payments computing systems of the payments computing systems. 
     
     
         17 . The method of  claim 15 , further comprising retraining the model based on aggregated gradient updates received from at least two payments computing systems of the payments computing systems. 
     
     
         18 . The method of  claim 15 , wherein at least one payments system of the plurality of payments computing system communicates via an enterprise computing network local to the computing platform. 
     
     
         19 . The method of  claim 15 , wherein at least one payments system of the plurality of payments computing system communicates via an external computing network communicatively coupled to the network and wherein the network is local to the computing platform. 
     
     
         20 . The method of  claim 15 , wherein the first payments computing system operates using a first messaging protocol and a second payments computing system operates using a second messaging protocol.